PSV3_1

[Rocky640]

Small speed-up in pawn.cpp
No functional change
Results for 10 tests for each version:

Base Test Diff
Mean 1435636 1445238 -9602
StDev 22576 23189 1848

p-value: 1
speedup: 0.007

[locutus2]

In this form this is a functional change because you use other backward pawn values

[Rocky640]

In the current master code, a "backward" pawn is ALWAYS an "unsupported" pawn (since we exclude connected pawns)

Here I changed "backward" penalties for ("backward" + "unsupported" S(20,10) )
and changed the logic to avoid the "unsupported" addition again.

So whatever bench you can think of, it will ALWAYS give the same result.

This is also consistent with what was done with the Isolated pawns (the unsupported weight was already included)
This also has the merit to illustrate more clearly that the overall penalty for a backward pawn is greater than for an Isolated pawn, which I find rather surprising especially in the opposed case. I'm currently doing some local tests on those weights.
Finally, it groups all the "unsupported" cases together, (isolated, backward, and unsupported) which opens the way for some eventual simplifications.

[mcostalba]

I confirm the small speed up on mingw gcc 4.9.2 Results for 10 tests for each version: BaseTestDiffMean14578331460499-2666StDev74548751032606p-value: 84,70% speedup: 0,20%

[hxim]

My results:
Results for 20 tests for each version:

        Base      Test      Diff      
Mean    1279453   1284523   -5070     
StDev   1721      3313      2615      

p-value: 0,974

[glinscott]

Cool, looks good.

[Rocky640]

A simplification test is about to be run.
I suggest to wait for the result before committing this small speed-up.

[mstembera]

Maybe too late but simplifications should always run against the best current known version. So applying this first and then running the simplification against this would have been better.

[zamar]

approved. This patch makes sense logically.